Big AI vendors are racing to recoup massive investments by selling premium versions of chatbots and agent-based tools, but setting a price is proving elusive. The core unit of cost—tokens used to process prompts and generate outputs—has grown cheaper, even as consumption soars and results remain non‑deterministic, making value hard to pin down. Goldman Sachs projects monthly token use will surge 24-fold between 2026 and 2030 to roughly 120 quadrillion as companies adopt AI agents.
Enterprises are feeling the pinch. Firms experimenting internally often blow through token budgets; Microsoft has reportedly curbed some third‑party coding tools, and Uber burned through an annual allocation in months. “Trying to tie someone into a cost model… doesn’t make any sense,” says Saviynt’s Simon Gooch, citing fast‑moving token economics. Smaller companies exploit flat‑fee personal accounts—an arrangement observers expect platforms to clamp down on under shareholder pressure.
Vendors are weighing price hikes, outcome‑based fees or incident bundles, but any scheme can be upended by sudden changes in model pricing. “You get into variable pricing, and it’s changing every couple of months,” says Sumo Logic’s Bill Peterson. For now, companies must tighten prompts, pick models carefully, and accept that while more tokens can yield better results, they also complicate budgets.
Related article:
